搜索
您的当前位置:首页正文

Performance of an Intrusion Detection System under Different Techniques

来源:二三娱乐
Journal ofCommunication and Computer 12(2Ol5、146.154 doi:10.17265/1548—7709/2015.03.007 IN Performance of an Intrusion Detection System under Diferent Techniques Sadeq A1Hamouz Head ofCompumr Information Systems Department,Faculty ofInformation Technology,Middle East Universiyt,Amman 11831 Jordan Abstract:Nowadays,with the rapid growth in technologies,which depend on computers systems and networksthreats are also ,increasing enormously.So,a huge number of approaches have been developed to protect systems and networks and to increase the security since it is an essential requirement in the majoriyt of the applications.In this paper,a statistical Naive Bayesian method is applied in all IDS system using different scenarios.The performance of the IDS is measured through experiments using NSL..KDD dataset. Keywords:Intrusion detection system,attack trends,securiy,atttacker,techniques,network threats,network defense 1.Introduction Previously,severa1 researchers have described the design of IDSs to offer details and descriptions of the main characteristics of those systems that are applicable and relevant in the detection of attacks.The detector is a combination of both the a.box and d—box. The IDS characteristics can be represented using two relatively simple parameters.The first parameter indicates a general characteristic of the system,such as the capability to concern accepted expression design of IDSs depends on the experiences,which are resulted from the improvement and use of those systems in various fields and on the analysis of several matching on information.However,this parameter has not the ability to identify the scope where it is obtainable.In other words,it cannot discover the type of information expression matching that should be types ofattacks[1]. Some of IDS characteristies are:information,which used in the analysis,the level of the interpretation and verification of protocols and methods used in concerned.The second parameter has the ability to discover the IDS scope,which determines the validity of the system characteristics[4]. The IDS scope,which is an iterative method that consists of three high—level scopes,is explored.Those scopes are networking, user and host. Both discovering activities that may signify aRacks.Those systems range from simple to complex ones and differ in their characteristics【2]. In Ref.[1】,a model of IDS that consists of sensors and detectors was described.The used sensors to networking and host are divided into several low-level scopes,like application layer and process,while the signify the e-box of the CIDFA(Common Intrusion Detection Framework Architecture)and to recover information from a data source were described in Ref user scope is the human who uses the Ins[4].Several works about the analysis of IDSs were published in order to detect aRacks.Those works classify the IDSs, ID and ̄tacks.0ne of those works is the MAFTIA 『31.This recovered information is inserted then to the detector.Researchers in Ref.[1】explored that the C0rresp0nding author:Sadeq AlHamou ̄Ph.D.,research project that uses several concepts,models and terminologies that derived from reliable ifelds[5,6]. Network threats can be persons,events or objects, which can cause damages in a network.Threats also ifelds:Internet,E—business and E-government.E-mail: sadeqhamouz@gmail.com. Performance of an Intrusion Detection System under Diferent Techniques 147 can be accidental,such as errors in calculations or malicious,such as data intended modification. Network security threats are divided into two main types:intemal and external threats.The internal threats happen by a person who has a pre—defined access to the network.This access can be an account or physical access.On the other hand,the extemal threats happen by persons who have no pre-defined accesses to the network.Those threats are resulted from the internet or access servers[7,8]. There are several types of attack trends,such as: threat activity trends,vulnerability trends,malicious code trends,fraud activity trends and phishing activiyt trends.Vulnerabiliyt is the weakness in a network that allows aRacker to cooperate the accessibility and integrity of this network.Malicious code is a wide group of software threats that attack networks and systems.The most complicated threats types are obtainable by the malicious code,which uses vulnerabilities in networks.Any code that changes, obliterates or takes data permits illegal access, damages a network and/or results in severa1 thins unrelated for users『9]. Phishing represents data from persons,groups or organizations with the use of a speciifc brand. Phishing attackers get several sensitive personal data of users.They need fatalities in order to offer their main qualiifcations.Fraud is the unauthorized or illegal use by attackers of some data that are related to a speciifed person【10]. Network defence is the actions taken to monitor, protect,analyse,detect,and respond to unauthorized activities in information systems and computer networks.Several systems are used in the defence of networks against attacks.The first method of protection is the IDS.IDSs have the ability to detect several types of aRacks by monitoring networks. Another protection method is the firewall,which is one of the most used defence devices that range from persormel ifrewall to array ones.Firewalls are utilized in the protection of large networks in large organizations.They are utilized to distinguish between networks via utilizing several rules in order to decide the allowable connections.Another protection method is the encryption,which is used to hide data using a secret algorithm.Those data are then decrypted only by a pre-defined secret key.In this way,attackers cannot reach these data.Another defence method is the authentication,which is similar to the encryption one.In this technique,messages are sent between a client and network access router by a protocol as a carrier in an authenticated way where attackers cannot reach those messages.After the authentication process, the client is defined as a MAC(media access contro1) address that can access the network and an AP(access point),which is defined also as a MAC address with the same client[1 1,12]. The last defence method is the ph) sical security, which assists in the evaluation and understanding of several risks which in turn facilitates taking corrective actions.It is the physical protection level that surrounds the neighbouring the intended coverage region with the proposed level of securiyt as well as threat model【13]. General introduction of intrusion detection systems is discussed;Section 2 offers an overview of IDSs. Section 3 discusses the IDS classification and filtering Section 4 analyses the system model of IDS using Naive technique,the evaluation and the comparison between the proposed systems section with different features are presented in Section 5 and conclusion is given in Section 6. 2.Background IDSs are divided into two groups:NIDS(Network IDS、and HIDS(Host IDS).The NIDS monitors the behaviour of hte system.while the HIDS monitors the calls of the system.For the NIDS.the activities of the network are independent on severa1 ports.Random Drojection sketches are used in order to decrease the dimensionaliyt of information with the use of multi..resolution non..Gaussian marginal distribution in 148 Performance of an Intrusion Detection System under Diferent Techniques order to find out the abnormalities across several levels of aggregation[14,1 5].The entropy based method was used in the whole network trafifc[1 6]. Both the statistical tests and the subspace techniques, which suppose that the connection features are normally distirbuted,were used[1 7,1 8】. NIDS are widely used as the last defence 1ine in order to allow several event responses when the intrusion avoidance mechanisms are not effective. This system compares the network trafifc with a known database in order to detect the unwanted trafifcs.The main benefits of NIDS are:its ease use and few numbers of generated false alarms.In contrast, NlDS cannot detect all types of attacks in an effeclive way.Some ofthose aRacks are:U2R,I L【7]. The most common ytpes of attacks are"DOS,U2R, I L and probe attacks【19]. DoS(Denial of service)attackers use obtainable or unobtainable memory sources in order to control requirements or to ignore rights of users from service, some of those attackers are SYN lfood,neptune,back, smurf,land and teardrop. U2R ruser to root)attackers use an account of a system user in order to realize root access to the required system as the user privilege(e.g.,buffer overflow). R2L(remote to loca1)attackers send severa1 packets to the system without having an account on this system(e.g.,password guessing)・ Probe attackers find out information or recognized threats.Attackers can easily make an attack with the use of this information(e.g.,ping sweep,port scan). The HIDS works on discovering the system calls. Those works are divided into two groups:sequence based works and feature based works.The sequence based works depend on the events chronological orders,while the second works consider the calls as independent information elements.In Re£[20], information was simpliifed by creating a database storing calls subsequence and then examined them.In RLef.【81,the richer group ofattributes was assumed to be a return value where influences are related to the system calls. HIDS is used in the detection of intrusions via examining several computing activities models,like the CPU usage and memory.HIDS analyses the system settings,system calls,local log inspections and more.It is used widely due to its effectiveness in the detection of known attacks.On the other hand,this system is not effective in the detection of new attacks 【211. Both NIDS and HIDS differ rfom each other,but in the same time,both are complement to each other.In other words,a real secured environment needs the use ofboth systems in order to offer a forceful system htat is considered as the foundation for monitoring and detecting misuses.This combination can filter alerts as well as notifications in a perfect way,which in turn helps in controlling and reacting to misuses. The main three types of detection methodologies are:pattem matching,protocol analysis and anomaly detection. The pattern matching methodology is used to determine how frequently an applicant pattern happens and also to determine some data about its rfequency distribution throughout a text.PaRem is set of strings,in which each string is a series of symbols. The best pattem has a small number of strings.This technique depends on finding out how many times a string is occurred in a text and determining its incident positions[22]. The protocol analysis technique is used to discover the locations as well as lengths of fields that exist in the protocol packets.The structure of both needs and responses can be understood by these packets with the use of reverse engineering.This technique is carried out via hand using perception as well as a protocol analyser instrument,like tcpdump.It can be used in the NIDSs in order to find out the higher—level semanticffamevmrkfrom atrafifc set[19,231. The anomaly detection technique is used in order to discover pattems in data,which are not matched with Performance of an Intrusion Detection System under Diferent Techniques l49 the prospected behaviour.Those patterns can be anomalies,exceptions,contaminants,peculiarities or outliers.The most used patterns in this technique are the anomalies and outliers.This technique can discover wide applications use,like fraud detection of credit cards and intrusion detection.The most important point of this technique is that anomalies that exist in data are converted into important data in several applications【1 6,241. 3.Multi—Layer Bayesian Filtering Technique Multi-layer Bayesian filtering technique is used in the IDSs with the use of KDD dataset.KDD is one of the main practical and realistic sets that contain actual attacks.It is used in the modelling and evaluation of IDS and it assists in the comparison between experimental results.The model of Bayesian IDS identifies features,which have diverse happening probabilities in both attacks and TCP trafifc.Initially, Bayesina filter is qualiifed by a pre—classiifed trafifc and then it corrects the features probabilities.After that, it calculates each TCP probability and categorizes it as a normal trafifc or attack one. Bayesian filter contains two components:training engine and testing engine.The training engine calculates the numbers of both good and bad records, then it creates three hash tables;two of them contain the frequency of both good records attributes and bad records attributes and the third one contains those attributes and hteir scores[9,1 3]. The testing engine is used to test the resultant training engine using the KDD dataset and to determine if the record is an attack or not based on a speciifed threshold.Accuracy and results of tests depend on databases,features and trheshold value. The following percentage expressions are used in the analysis ofdata[10,7,25]. TN(true negative):normal records,which are correctly classiifed;TP(true positive):attack records, which are correctly classiifed;FP(false positive): normal records,which are incorrectly classified as aRacks and FN(false negative):attack records which are incorrectly classified as norma1.By using these expressions,both the detection rate and classification rate can be represented as follows: ’ DetectionRate(DR) T D (1) Classiifcation Rate(cR)= T P+ TN (2) Three improved Bayesian filters are explained and compared with each other in order to determine the most accurate iflter in the detection of attack records. Improved Bayesian Filter 1(IBF 1) IBF 1 is a one—layer filter.In this filter,the normal records are filtered again for several times with the use of engines that have different settings of trheshold and features,where the output normal records of one engine are the inputs of the next engine.This process enhances the accuracy,where attack records of all engines are collected.Fig.1 shows the IBF 1 model [12,26]. Improved Bayesian Filter 2(IBF2) IBF2 is a one—layer filter.In this filter,the normal records are filtered again for several times with the use of engines that have different settings of databases, where hte output normal records of one engine are the inputs of the next engine.This process enhances the detection rate but it has high FP percentage which is a problem.Fig.2 below shows the IBF2 model[1 1]. Improved Bayesian Filter 3(IBF3) IBF3 filter consists of two layers in series.In the ifrst layer,attack records are filtered again,while in the second layer,both attack and normal records are ifltered.This combination gives the highest detection rate that equals to 96.85%since both types of records are filtered again.Fig.3 below shows the IBF3 model [14,17]. 4.System Model In this paper,a Naive Bayesian based IDS is explored and discussed. Initially,this section offers a briefdescription conceming the Naive Bayesian 15O Performance of an Intrusion Detection System under Diferent Techniques Fig.1 IBF1 mode1. Fig.2 IBF2 mode1. technique with exploring their principles of work and main equations.Atier that,those systems are trained this section.The NSL—KDD database iS applied in this work using two sets of features numbers for both and tested using the NSL—KDD database in order to measure and evaluate their performance.This database composed of 4 1 network connection features, where the names of those features are demonstrated in classiifcation methods:(5,10,24,29,33,34,38,40) and(2,5,8,23,30,34,35,38).The proposed classification methods are applied on the proposed IDSs using those sets of features. Performance of an Intrusion Detection System under Diferent Techniques 151 Fig.3 IBF3 mode1. The Naive Bayes algorithm is applied on the IDS to NSL—KDD data sets include 4 l features of the ifnd the probability of the presence of an attack in a network connection. computer network.When the computed attack probability is high but not enough to be considered as 5.Simulation and Numerieal Results attack,then the computer network produces a report In mis section.the obtained results for developed and wams the administrator of hte system.The Naive Naive Bayes based IDSs(intrusion detection systems、 Bayes can be used to classify any unknown object using hte proposed sets offeatures ni Refs.[27,28】are when the network is quantified based on its attributes presented.The NSL—KDD database iS used to measure values by using the following expression Eq.(3)[15]. the performance of both systems,where those systems P(Cv(x/cj)P(Cj)differ in the used set of features.A11 the simulation j/x)=— 一  results are obtained using the MATLAB program. P(Cj/X)>P(Ci/X),1 i m,i≠j (3) Fig.4 shows the obtained results of hte proposed Where,C{class that belong to group of m classes C 1, Naive Bayes based IDS using the proposed set of C2...Cm,X represents the data sample which not featuresinRef f271 forthe first case. known and P㈣is constant for each category.The As shown in the Fig.5 below.the system has DR proposed IDSs use Naive Bayes Eq.(3)to classify 88.5 l%and obvious FN and FP rates.where the network connections as normal or attack based on resultant FN rate in 7.49%and FP rate iS l8.16%. their features.In the proposed IDS hte NSL—KDD data Thus,this system needs further enhancements based set is used ofr rtaining and testing stages to evaluation. on eliminating all records that result in false alarms. Performance of an Intrusion Detection System under Diferent Techniques 153 IDS using Naive Bayesian classiifer to decrease the number of generated false alarms:false positives and false negatives,improve the network security and enhance the detection rate of several types of aRacks. Naive Bayesian method was applied on the constructed IDS in different scenarios using the MATLAB program,where then a comparative study among them was conducted based on analyzing the performance parameters and determining the most efifcient statistical method in detecting various types of attacks.The NSL—KDD database was used to measure the performance of the implemented systems, where it composed of 4 1 features of the network connection. For the Naive Bayesian classifier,two systems have been implemented and analyzed.Using the proposed features numbers in Refs.[27,28],the obtained DRs are 88.51%and 87-2%.FP rates are 18.16%and 15.11% and FN rates are 7.49%and 10.38%,respectively. References 【l】Dacier,M.,andAlessandri,D. 1 999.“VulDa:A Vulnerability Database.’’ 『21 Alessandri,D.2004.“Attack-Class-Based Analysis of Intrusion Detection Systems.”A thesis submitted t0 School of Computing Science。University of Newcastle upon Tyne. 【3】 xu,J.,and Shelton,C.R.20 1 0.“Intrusion Detection Using Continuous Time Bayesian Networks.’’Journal of Arffifciaf Intelligence Research39:745—74. 『41 MAFTIA Consortium. “Towards a Taxonomy of Intrusion Detection Systems and Aaacks.”D.Alessandri. ed.Malicious.and Accidenta1.Fault Tolerance for Internet Applications. [5】Dewaele,G.,Fukuda,K.,and Borgnat,P.2007. ‘‘Extracting Hidden Anomalies Using Sketch and Non-.Gaussian Multi--resolution Statistical Detection Procedures.”In Proceedings ofthe 2007 Workshop on Large Seale Attack Defence.145—52. 【6】Lakhina,A.,Crovella,M.,and Diot,C.2005.“Mining Anomalies Using Tra币c Feature Distributions.”AC sIGCOMM Computer Communication Review . Proceedings of the 2005 Conference on Applications, Technologies,Architectures,and Protocols for Computer Communications 35:21 7.28. 『71 Federal Information Processing Standards Publication l9l(FIPS PUB l91 .“Guideline for the Analysis Local Area Network Security”. 【8】Heidari,M.,2004.Malicious Code in Depth,pp.1・2 1. 【9】 Financial Crimes Enforcement Network(FCEN), “Mo ̄gage Loan Fraud”.2008. [10]Raman,B.2005.Cryptography and Network Security, Department of CSE,lIT Kanpur. 『l1 1 Faria,D.2006.Scalable Location—Based Security in Wireless Networks khan.A dissertation submitted to the Department of Computer Science and the Committee on Graduate Studies of Stanford University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy. 【12]Gogoi,P.,Bhattacharyya,D.K.,Borah,B.,and Kalita,J. K.2013.“MLH—IDS:A Multi.Level Hybrid Intrusion Detection Method.’’The Computer Journal Advance Access published. 【13】Aickelin,U.,Twycross,J.,and Roberts,T.H.2007. “Rule Generalization in Intrusion Detection Systems Using SNORT.’’International Journal of Electronic Securiyt andDigitalForensics:101.16. 【14】Hu, J. 2010.Host-Based Anomaly Intrusion Detection.Springer. 【15]Navarro,G.2010.“Pattern Matching.” e Computer Journaf Advance Access published 58:1.24. 【16】Beddoe,M.A.2005.‘'Network Protocol Analysis Using Bioinformatics Algorithms.’’In Proceedings of the Seventh International Network Conference. [1 7】Dreger,H.,Felmann,A.,Mai,M.,Paxson,V.,and Sommer,R.2006.“Dynamic Application-Layer Protocol Analysis for Network Intrusion Detection.’’ In UsENIX-ss’o6 Proceedings of the 15th Conference on UsENIX Security Symposium [1 8】Montgomery,D.C.2001.Introduction to Statistical Quality Contro1.4th ed.,New York:Wiley,NY. 【19】Hofmeyr,A.,Forrest,S.,and Somayaji,A.1998. “Intrusion Detection Using Sequences of System Calls.’’ JournalD厂ComputerSecurity6:151.80. 【20】Leung,J.2008.“Vulnerabilivy Management—A Guide to Managing Intemal and External Threats.” [2 1】Tandon,G.,and Chan,P.K.2005.“Modelling Multiple Time Series for Anomaly Detection.’’Presented at the Florida Artificia1 Intelligence Research Society Conference. 【22]Ye,N.,Emran,S.M.,Chen,Q.,and Vilbert,S.2002. “Multivariate Statistica1 Analysis of Audit Trails for Host—based Intrusion Detection.’’Transactions of Computers 51:810—20. 【23】Pporras,P.,Schnacenberg,D.,Chen,S.S.,Stillman,M., nad Wu.F.1 998.“The Common Intrusion Detection Framework Architecture.’’ 【24】Sun,Z.,Kaucic,R.,Mendoca,P.,and Can,A.2007.“A Statistical Approach to Industrial AnomalY Detection.’’ 】.9. 154 Performance of an Intrusion Detection System under Diferent Techniques [25] Altwajry,H.,and Algarny,S.20 1 1.‘'Multi-Layer Bayesian Based Intrusion Detection System.” In Features Selection for Detecting Network Intrusions.’’ Presented at the Intemational Conference for Intemet Proceedings ofthe Worm Congress on Engineering and Compu ̄r Science. Technology and Secured Transactions ICITST. 【28]Bhoria,P.,and Garg,K.20 1 3.“Determining Feature Set of DOS Attacks.”international Journal of Advanced Research in Computer Science and Software l ̄rngmeering. ,and Kumar,V.2009.“Anomaly Derection: 【26】 Chandola,V.A Survey.’’ACMCompu ̄ngSurveys(CSUR)41:1-72. ,W.,A.,and Naoum,R.2009.‘‘Significant of 【27] Sharafat

因篇幅问题不能全部显示,请点此查看更多更全内容

Top